Re: [PATCH 5/6] mm: honor PF_MEMALLOC_NOMOVABLE for all allocations

[Pavel Tatashin <pasha.tatashin@soleen.com>]

On Thu, Dec 3, 2020 at 3:17 AM John Hubbard <jhubbard@nvidia.com> wrote:
>
> On 12/1/20 9:23 PM, Pavel Tatashin wrote:
> > PF_MEMALLOC_NOMOVABLE is only honored for CMA allocations, extend
> > this flag to work for any allocations by removing __GFP_MOVABLE from
> > gfp_mask when this flag is passed in the current context, thus
> > prohibiting allocations from ZONE_MOVABLE.
> >
> > Signed-off-by: Pavel Tatashin <pasha.tatashin@soleen.com>
> > ---
> >   mm/hugetlb.c    |  2 +-
> >   mm/page_alloc.c | 26 ++++++++++++++++----------
> >   2 files changed, 17 insertions(+), 11 deletions(-)
> >
> > diff --git a/mm/hugetlb.c b/mm/hugetlb.c
> > index 02213c74ed6b..00e786201d8b 100644
> > --- a/mm/hugetlb.c
> > +++ b/mm/hugetlb.c
> > @@ -1036,7 +1036,7 @@ static struct page *dequeue_huge_page_node_exact(struct hstate *h, int nid)
> >       bool nomovable = !!(current->flags & PF_MEMALLOC_NOMOVABLE);
> >
> >       list_for_each_entry(page, &h->hugepage_freelists[nid], lru) {
> > -             if (nomovable && is_migrate_cma_page(page))
> > +             if (nomovable && is_migrate_movable(get_pageblock_migratetype(page)))
>
>
> I wonder if we should add a helper, like is_migrate_cma_page(), that avoids having
> to call get_pageblock_migratetype() at all of the callsites?

Good idea, I will add it.

>
>
> >                       continue;
> >
> >               if (PageHWPoison(page))
> > diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> > index 611799c72da5..7a6d86d0bc5f 100644
> > --- a/mm/page_alloc.c
> > +++ b/mm/page_alloc.c
> > @@ -3766,20 +3766,25 @@ alloc_flags_nofragment(struct zone *zone, gfp_t gfp_mask)
> >       return alloc_flags;
> >   }
> >
> > -static inline unsigned int current_alloc_flags(gfp_t gfp_mask,
> > -                                     unsigned int alloc_flags)
> > +static inline unsigned int cma_alloc_flags(gfp_t gfp_mask,
> > +                                        unsigned int alloc_flags)
>
> Actually, maybe the original name should be left intact. This handles current alloc
> flags, which right now happen to only cover CMA flags, so the original name seems
> accurate, right?

The reason I re-named it is because we do not access current context
anymore, only use gfp_mask to get cma flag.
>> -     unsigned int pflags = current->flags;

So, keeping "current" in the function name makes its intent misleading.

Thank you,
Pasha